Impact of COVID-19 on epidemic trend of hepatitis C in Henan … – BMC Infectious Diseases

A total of 267,968 cases of hepatitis C were diagnosed between January 2013 and September 2022, resulting in an annual average diagnosis rate of 2.42 per 100,000 people. The highest diagnosis rate occurred in 2013 at 2.97 per 100,000 people, while the lowest was recorded in 2020 at 1.7 per 100,000 people. The average monthly diagnosis cases of hepatitis C was 2291.

The diagnosis rate for hepatitis C was analyzed using the ARIMA model. Figure2 displays a time series analysis of hepatitis C diagnosis rate in Henan spanning from January 2013 to September 2022. The data is divided into two segments: pre-COVID-19 (from January 2013 to December 2019) and post-COVID-19 (from January 2020 to September 2022). The insights gained from Fig.2 highlight a noticeable decrease in hepatitis C diagnosis rate during the COVID-19 outbreak in 2020, followed by a subsequent increase. Across the broader timeframe from 2013 to 2022, the diagnosis series of hepatitis C exhibited cyclic patterns. Through the application of differencing to adjust for this trend and periodicity, the hepatitis C diagnosis series demonstrated enhanced stability.

Henan Province Hepatitis C Time Series Plot

The Box-Cox method was employed, along with the auto.arima function in the R software, to fit the hepatitis C diagnosis series in Henan Province from January 2013 to September 2022. ACF and PACF plots are depicted in Supplementary Fig.S1. In this illustration, bars above or below the dotted line represent statistically significant autocorrelation (P<0.05). Both ACF and PACF plots (Supplementary Fig.S1a) reveal undifferentiated autocorrelation and partial autocorrelation patterns, with noticeable significant autocorrelation. Supplementary Fig.S1b presents autocorrelation and partial autocorrelation post-differencing. In comparison to Supplementary Fig.S1a, differencing effectively removed much of the autocorrelation.

The ARIMA (0,1,1) (0,1,1)12 model yielded the lowest AIC (1459.58), AICc (1460.19), and BIC (1472.8). Consequently, this model structure was selected as the optimal. Diagnostic results indicated that for the ARIMA (0,1,1) (0,1,1)12 model, MA1=-0.62 (t=-8.06, P<0.001) and SMA1=-0.79 (t=-6.76, P<0.001); ACF and PACF plots of residuals indicated most correlation coefficients were within the confidence interval (Supplementary Fig.S2). The Ljung-Box Q test results demonstrated no statistically significant difference among residuals for different lag periods (P=0.408), affirming that model residuals constituted white noise. These results validate the ARIMA (0,1,1) (0,1,1)12 model.

Over time, the time series plot displayed relatively constant variance. The histogram of the time series showed normally distributed prediction errors, with the mean adhering to normal distribution as well. Residuals showcased no discernible pattern or significant autocorrelation, and they followed a normal distribution. The Ljung-Box Q test yielded a P-value of 0.408, confirming the chosen models good fit.

The final model estimated a step change of -800.0 (95% CI -1179.9 ~ -420.1, P<0.05) and a pulse change of 463.40 (95% CI 191.7~735.1, P<0.05) per month. Figure3; Table1 depict the comparison between predicted and observed values of the ARIMA model without intervention (counterfactual).

From January 2013 to September 2022, an analysis was conducted on the fitting and observed values of hepatitis C diagnosis cases both before and after the onset of COVID-19. The findings suggested a decrease in hepatitis C diagnosis cases in January 2020, with an anticipation of the impact being transient. Since January 2020, the influence of COVID-19 on hepatitis C diagnosis rate was modeled using a pulse function. Following the emergence of COVID-19, a decline in hepatitis C diagnosis cases was observed. A step function was utilized to fit potential long-term fluctuations in the number of hepatitis C diagnosis cases. The final model revealed a sudden reduction in diagnosis cases post the onset of COVID-19, followed by a gradual increase back to pre-COVID-19 levels. Given the nature of the intervention, we assumed an immediate decrease in diagnosis cases post-intervention (step change) and an accompanying pulse change. Hence, variables representing both types of impacts were incorporated into the model.

Moreover, when comparing the ITS-ARIMA model with the BSTS model (Fig.3), results indicated that the mean absolute percentage error (MAPE=19.95%) of the ITS-ARIMA models predictions was lower than that of the BSTS model (MAPE=25.7%). This implies that the prediction performance of the ITS-ARIMA model surpassed that of the BSTS model, demonstrating the formers superior predictive capabilities.

Observed and predicted values without intervention based on ITS-ARIMA model

Read more:

Impact of COVID-19 on epidemic trend of hepatitis C in Henan ... - BMC Infectious Diseases